Which AI model should I use?
Match cognitive load to model capability
The Decision Framework
Three questions to select the right model:
| Question | If Yes | Model Tier |
|---|---|---|
| Is this execution? | Well-defined task, clear success criteria, checklist exists | Small Model (no thinking) |
| Does this need judgment? | Guidelines exist but interpretation needed, some ambiguity | Mid Model + Thinking |
| Is this complex? | Define the problem then solve it, strategic reasoning required | Top Model + Thinking |
Context Modifier: Skills, Projects, or Instructions with domain knowledge reduce complexity by one tier.
Cross-Platform Model Reference
| Task Type | Claude | OpenAI | Thinking? | |
|---|---|---|---|---|
| Structured / Execution | Haiku | Instant | Flash | No |
| Unstructured / Judgment | Sonnet | Auto | Thinking | Yes |
| Complex / Strategic | Opus | Thinking | Pro / Ultra | Yes |
The Employee Analogy
| Employee Level | Task Type | What They Do |
|---|---|---|
| Junior | Execution | Executes instructions from a checklist |
| Mid-Level | Judgment | Interprets intent within guidelines |
| Senior | Complex | Frames the problem, then solves it |
Don't pay senior rates for junior work. Don't expect junior output on senior problems.
Practical Examples
| Task | Model | Why |
|---|---|---|
| Format a document, run a template | Small | Execution |
| Site audit, data extraction | Small | Execution |
| Evaluate my LinkedIn headline | Mid + Thinking | Judgment |
| Debug code, implement feature | Mid + Thinking | Judgment |
| Blindspot analysis, strategy development | Top + Thinking | Complex |
| System architecture, novel frameworks | Top + Thinking | Complex |
Key Rules
- 1. Always enable thinking mode for Mid and Top tier models
- 2. Context (Skills / Projects / Instructions) drops complexity by one tier
- 3. Pay compute upfront to avoid correction tax downstream
- 4. The less structured the request, the more advanced the model
Compound Leverage | THINK Methodology | compoundleverage.com